GreenCard Radar

Accuracy Scorecard

Most Visa Bulletin prediction tools give you a forecast, then disclaim that it “may vary significantly” — without ever showing how often they were actually right. We publish a per-category hit rate, updated as each bulletin lands — and we count the misses.

1-month forecast
75%
within ±31d · median miss 7d
3-month forecast
44%
within ±31d · median miss 43d
6-month forecast
26%
within ±31d · median miss 102d

Hit rate across 12,197 backtested one-month projections (all categories, both charts). Accuracy decays sharply with horizon — a one-month call is reliable, a six-month one is a rough guide. We show both rather than advertise only the flattering number. “Median miss” is the typical error; the mean is larger because a few retrogressions miss badly.

How it’s computed

We hold a complete, committed series of 202 consecutive monthly bulletins (October 2009July 2026). We replay the exact velocity model the site ships — trailing 24-month median pace — as it would have run in every past month, then compare its projection to what the State Department actually published 1, 3, 6 months later. A projection is a hit if it lands within ±31days of the real cutoff. Months where a category went Current or Unavailable aren’t date predictions, so they are reported separately and never counted in the denominator.

Each cell is graded green ≥70%, amber 50–69%, red <50% — the red ones stay visible.

By category & country

CategoryChart1mo3mo6moMedian miss (1mo)
EB-2 — IndiaFinal Action Dates52%25%13%28d
EB-2 — ChinaFinal Action Dates63%32%9%23d
EB-3 — IndiaFinal Action Dates76%39%28%12d
EB-3 — ChinaFinal Action Dates54%17%5%31d
EB-2 — IndiaDates for Filing84%61%41%0d
EB-3 — IndiaDates for Filing84%60%36%0d
EB-5 Unreserved — ChinaDates for Filing91%73%58%0d
F2A — MexicoFinal Action Dates53%34%22%30d
F3 — MexicoFinal Action Dates90%63%39%7d
F4 — MexicoFinal Action Dates92%74%53%7d
F4 — PhilippinesFinal Action Dates70%37%27%17d

Final Action Dates govern green-card approval; Dates for Filing govern when you may submit. Filing dates move more predictably, which is why their scores run higher. Full per-cell data (all 116 categories) is in the build artifact behind this page.

How a live prediction is logged

The table above is a historical backtest. Going forward, each live forecast is also written to an append-only log before the corresponding bulletin is published, then reconciled against the actual cutoff once it appears — the record cannot be edited after the outcome is known. See the methodology for the model itself.

Sources & verification

Compiled and reviewed by the GreenCard Radar editorial team · Data as of July 2026

How we build & backtest these estimates →